The news: Meta’s Llama 4 is tackling “left-leaning bias” and removing guardrails against contentious political topics.

• The company stated on its blog that Llama 4 refuses less than 2% of queries on debated political and social issues, compared with 7% for Llama 3.3.
• Meta said that the prompts it now refuses are “dramatically more balanced” and that its responses have a comparable political lean to xAI’s Grok.